摘要
ThisletterpresentsanewchunkingmethodbasedonMaximumEntropy(ME)modelwithN-foldtemplatecorrectionmodel.Firsttwotypesofmachinelearningmodelsaredescribed.Basedontheanalysisofthetwomodels,thenthechunkingmodelwhichcombinestheprofitsofconditionalprobabilitymodelandrulebasedmodelisproposed.Theselectionoffeaturesandruletemplatesinthechunkingmodelisdiscussed.ExperimentalresultsfortheCoNLL-2000corpusshowthatthisapproachachievesimpressiveaccuracyintermsoftheF-score:92.93%.ComparedwiththeMEmodelandMEMarkovmodel,thenewchunkingmodelachievesbetterperformance.
出版日期
2007年05月15日(中国期刊网平台首次上网日期,不代表论文的发表时间)